CN106851571A - WiFi localization methods in a kind of quick KNN rooms based on decision tree - Google Patents

Abstract

The invention discloses WiFi localization methods in a kind of quick KNN rooms based on decision tree, methods described specifically includes following steps：Positioning region is divided into many sub-regions, multiple elements of a fix points are set in each sub-regions；Terminal gathers each coordinate points RSSI finger print informations and coordinate information, by wireless network transmissions to server, builds fingerprint database；Server is differentiated by integrated decision Tree algorithms to area classification residing for target；Use KNN algorithms to be matched with classification residing for target, calculate exact position；Positioning result is back into terminal to show.The indoor WiFi localization methods of the quick KNN based on decision tree of present invention design, overcome the slow problem of traditional KNN algorithms locating speed, territorial classification is carried out to positioning target using decision Tree algorithms, target is accurately positioned using KNN algorithms, localization method is all significantly improved in positioning precision and efficiency.

Description

WiFi localization methods in a kind of quick KNN rooms based on decision tree

Technical field

The present invention relates to communicate, Signal and Information Processing and location Based service technical field, and in particular to Yi Zhongji In WiFi localization methods in the quick KNN rooms of decision tree.

Background technology

With the fast development of mobile interchange mobile network, location Based service possesses the market with rapid growth, its Middle indoor positioning quickly grows in recent years.The application of positioning is generally to use global positioning system, but due to indoor environment without Method relies on the signal that gps satellite transmission comes, and indoor environment is usually relatively complex so that the positioning precision of indoor locating system It is a greater impact, which prevent the application of indoor locating system.Current various indoor positioning technologies Remarkable Progress On Electric Artificials enter Exhibition, wherein WiFi technology is to be applied to one of most technology in indoor positioning research field, and it has, and signal coverage rate is high, terminal Number of users is big and the features such as long transmission distance.

Most of alignment systems based on WiFi are all to carry out position mark using received signal strength (RSSI).It is based on The method of RSSI is largely divided into two classes：Triangle is positioned and location fingerprint recognizer.Triangle positioning be using signal distance- The target to be measured of loss model calculating arrives the distance between multiple known reference points information estimation final goal position, and location fingerprint Identification then derives target location by the signal characteristic finger print information of the RSSI and reference point that compare point to be determined.Triangle is determined Position is because indoor environment complexity is so that positioning result is unstable.

Location fingerprint localization method based on RSSI, generally comprises offline and online two stages.Off-line phase, first will Space is divided into latticed area distribution, and gathering finger print information in each reference point by mobile device sets up fingerprint base. Line stage reference point RSSI Vectors matchings then terminal in RSSI that unknown position is collected into vector with fingerprint base, by Final location estimation is carried out with algorithm.Typical pattern matching algorithm is KNN algorithms, and Euclidean distance is used in the algorithm For metric objective vector and the matching degree of sample vector.

However, due to needing to calculate the tested point RSSI vectorial Euclidean distances with whole fingerprint base when calculating similarity, When fingerprint database is huger, it may be desirable to spend longer time.

The content of the invention

The invention aims to solve drawbacks described above of the prior art, there is provided a kind of based on the quick of decision tree WiFi localization methods in KNN rooms, the method utilizes radio network technique and indoor fingerprint location technology, by determining in server Position algorithm is matched to data, is realized the quick identification of indoor regional area and is accurately positioned.

The purpose of the present invention can be reached by adopting the following technical scheme that：

A kind of WiFi localization methods in quick KNN rooms based on decision tree, methods described comprises the following steps：

Positioning region is divided into many sub-regions, multiple elements of a fix points are set in each sub-regions；

Terminal gathers each coordinate points RSSI finger print informations and coordinate information, by wireless network transmissions to server, structure Build fingerprint database Ψ；

Server is differentiated by integrated decision Tree algorithms to area classification residing for target；

Use KNN algorithms to being matched with classification residing for target, calculate exact position.

Further, it is described that positioning region is divided into many sub-regions, in each sub-regions, multiple positioning are set Coordinate points are specifically included：

Positioning region is carried out according to dividing mode at equal intervals divide many sub-regions, be that each sub-regions set classification Label；

Arbitrary placement's multiple elements of a fix point, records each point coordinates information in each sub-regions.

Further, described terminal gathers each coordinate points RSSI finger print informations and coordinate information, is encapsulated by JSON To pass through wireless network transmissions to server after network packet.

Further, described server carries out differentiation tool by integrated decision Tree algorithms to area classification residing for target Body includes：

To fingerprint database Ψ and label information, decision tree, the multiple leaf knots of generation are generated using decision tree training philosophy Point；

Input target sample enters decision tree root node, rule match is carried out with inner branch successively, until target sample Into leafy node；

Leafy node classification is determined that target sample affiliated area classification is leaf by inside comprising the most classification of sample number Node classification.

Further, described use KNN algorithms are matched to classification residing for target, are calculated exact position and are specifically wrapped Include：

Every cosine similarity of vector in the vectorial fingerprint bases corresponding with residing classification of RSSI of tested point is calculated, by carrying out Ascending order is arranged, and is taken preceding K reference point and is constituted neighbours' sample set, and the corresponding two-dimensional coordinate of neighbours' sample set constitutes neighbours' sample coordinate Collection；

Using the cosine similarity of neighbours' sample set as weight, tested point position coordinates is drawn using the method based on weighting (x,y)。

Further, described fingerprint database Ψ is expressed as：

Wherein RSSI_m,n(m=1,2...M, n=1,2 ... N) represent that m-th reference point receives n-th RSSI of AP Average value, each row vector of fingerprint database Ψ represents that a reference point receives the RSSI of N number of AP.

Further, described decision tree training philosophy is specifically included：

Regard RSSI vectors as a categorical attribute per one-dimensional component, therefore property set is expressed as：

R (D)={ R₁,...,R_i,...,R_N}

Wherein, R_iRSSI vector i-th dimension components are represented, for RSSI i-th dimension attributes R_i, to these values by from small to large Sequence, obtains ascending sequence { R_i1,...,R_ij,...R_in, set [R_ij,R_ij+1) intermediate pointIt is interval division point, For attribute R_iConstruction candidate divides point set：

Structure attribute optimum division point decision rule, i.e. attribute R_iOptimum division point should meet：

According to above-mentioned decision rule, optimal dividing point corresponding informance gain is attribute information gain in itself, in construction During decision tree, current node attribute should meet：

R=arg max G (D, R_i)

From root node, optimal dividing attribute and optimal dividing point are selected according to above-mentioned rule, by sample set according to drawing Branch carries out two points for two subsets, is then further divided in the two subsets, until all leafy nodes are all wrapped Sample containing identical category, completes decision tree and builds.

Further, in the vectorial fingerprint bases corresponding with residing classification of RSSI of described calculating tested point more than every vector String similarity is specific as follows：

Target sample r={ r₁,...r_N, residing category dataset each sample is designated as { (r_k1,...,r_ki,..., r_km, target sample and data set each sample cosine similarity is defined as：

Further, the method that described use is based on weighting show that tested point position coordinates (x, y) is as follows：

K maximum sample of similarity is chosen, is that each coordinate vector defines weight：

Point target positioning result to be measured is as follows：

Wherein, x_kiRepresent i-th coordinate vector abscissa of kth class sample, y_kiRepresent i-th coordinate of kth class sample Vectorial ordinate.

The present invention has the following advantages and effect relative to prior art：

(1) WiFi localization methods are effectively reduced because of indoor environment in the quick KNN rooms based on decision tree proposed by the present invention The more complicated and influence of the interference such as the multipath effect that causes and other signals.

(2) WiFi localization methods take full advantage of WiFi signal in the quick KNN rooms based on decision tree proposed by the present invention Coverage rate is high, infrastructure device disposes fairly perfect and long transmission distance advantage.

(3) WiFi localization method combination decision Tree algorithms in the quick KNN rooms based on decision tree proposed by the present invention, effectively The needs of problems of indoor area-of-interest positioning is solved, with conventional k-nearest neighbor, the calculation unlike SVMs scheduling algorithm Method effectively will be accurately positioned combination in region recognition and region.

(4) present invention is using WiFi localization methods in the quick KNN rooms based on decision tree and the WiFi based on other algorithms Localization method is compared, and due to having used the discriminant classification algorithm of decision tree in algorithm, discrimination reaches 90%；In positioning trip Between on, the fingerprint quantity of required matching during due to being accurately positioned narrowed down in identified region, so this method is determined Position efficiency will height compared to the localization method based on global fingerprint matching algorithm；In positioning precision, compared to comparative maturity KNN algorithms, positioning precision of the present invention is higher, and position error may remain in 1~2m.

Brief description of the drawings

Fig. 1 is experimental site region division schematic diagram, and its interior joint is exactly the reference point locations chosen；

Fig. 2 is that WiFi determines in the quick KNN rooms based on decision tree proposed for room area location requirement of the invention The flow chart of position algorithm.

Specific embodiment

To make the purpose, technical scheme and advantage of the embodiment of the present invention clearer, below in conjunction with the embodiment of the present invention In accompanying drawing, the technical scheme in the embodiment of the present invention is clearly and completely described, it is clear that described embodiment is A part of embodiment of the present invention, rather than whole embodiments.Based on the embodiment in the present invention, those of ordinary skill in the art The every other embodiment obtained under the premise of creative work is not made, belongs to the scope of protection of the invention.

Embodiment

The present embodiment is larger for indoor range and needs the Demand Design positioned to plurality of regional area A kind of localization method of the quick KNN based on decision tree.Judge which kind of region is target belong to using decision tree, with reference to weighting K Nearest neighbor algorithm calculates the exact position of target.

This example discloses WiFi indoor orientation methods in a kind of quick KNN rooms based on decision tree, process step figure ginseng According to shown in accompanying drawing 2, from accompanying drawing 2, the quick accurate indoor orientation method specifically includes following steps：

S1, positioning region is divided into many sub-regions, multiple elements of a fix points are set in each sub-regions；

In concrete application, step S1 is specially：

S101, positioning region is carried out according to dividing mode at equal intervals divide many sub-regions, be that each sub-regions set Put class label.

S102, arbitrary placement's multiple elements of a fix point in each sub-regions, record each point coordinates information.

S2, terminal gather each coordinate points RSSI finger print informations and coordinate information, by wireless network transmissions to server, Build fingerprint database Ψ；

In concrete application, step S2 is specially：

The RSSI information and coordinate information of terminal scanning each coordinate points, network packet is encapsulated as by JSON, is sent out It is sent to server.

Fingerprint database Ψ is expressed as：

Wherein RSSI_m,n(m=1,2...M, n=1,2 ... N) represent that m-th reference point receives n-th RSSI of AP Average value, each row vector of Ψ represents that a reference point receives the RSSI of N number of AP.

In positioning, terminal scanning WiFi signal obtains one group of RSSI fingerprint of positioning target, is inputted to positioning and calculates Method treatment.

S3, server are differentiated by integrated decision Tree algorithms to area classification residing for target；

In concrete application, step S3 is specially：

S301, to fingerprint database Ψ and label information, decision tree, the multiple leaves of generation are generated using decision tree training philosophy Child node.

In concrete application, the step S301 is specifically included：

S3011, regard RSSI vectors as a categorical attribute per one-dimensional component, therefore property set can be expressed as：

R (D)={ R₁,...,R_i,...,R_N}

Wherein, R_iRepresent RSSI vector i-th dimension components.For RSSI i-th dimension attributes R_i, to these values by from small to large Sequence, obtains ascending sequence { R_i1,...,R_ij,...R_in, set [R_ij,R_ij+1) intermediate pointIt is interval division point, For attribute R_i, it is possible to construct candidate and divide point set

S3012, structure attribute optimum division point decision rule, i.e. attribute R_iOptimum division point should meet：

S3013, according to above-mentioned decision rule, optimal dividing point corresponding informance gain is exactly attribute information gain in itself. When decision tree is constructed, current node attribute should meet：

R=arg max G (D, R_i)

From root node, optimal dividing attribute and optimal dividing point are selected according to above-mentioned rule, by sample set according to drawing Branch carries out two points for two subsets, is then further divided in the two subsets, until all leafy nodes are all wrapped Sample containing identical category, completes decision tree and builds.

S302, input target sample enter decision tree root node, rule match are carried out with inner branch successively, until target Sample enters leafy node.

S303, leafy node classification include the most classification of sample number and determine by inside, target sample affiliated area classification It is leafy node classification.

S4, classification residing for target is matched using KNN algorithms, calculate exact position；

In concrete application, the step S4 is specifically included：

Every cosine similarity of vector in S401, the vectorial fingerprint bases corresponding with residing classification of RSSI of calculating tested point, By ascending order arrangement is carried out, take preceding K reference point and constitute neighbours' sample set, the corresponding two-dimensional coordinate of neighbours' sample set constitutes neighbours' sample This coordinate set.

Target sample r={ r₁,...r_N, residing category dataset each sample is designated as { (r_k1,...,r_ki,..., r_km, target sample and data set each sample cosine similarity is defined as：

S402, using the cosine similarity of neighbours' sample set as weight, tested point position is drawn using the method based on weighting Put coordinate.

K maximum sample of similarity is chosen, is that each coordinate vector defines weight：

Point target positioning result to be measured is as follows：

Wherein, x_kiRepresent i-th coordinate vector abscissa of kth class sample, y_kiRepresent i-th coordinate of kth class sample Vectorial ordinate.

Finally, by follow-up information transmission, positioning result is back to terminal and is shown.

Embodiment two

This example is real by WiFi localization methods application in a kind of quick KNN rooms based on decision tree and experimental site region Floor area arrangement is tested as shown in figure 1, in the region of 10m*20m, 5 Wi-Fi hotspots are set altogether, is gathered with Android device RSSI fingerprints.

As Fig. 2 gives the flow chart that localization method is positioned, whole positioning process step is illustrated, in order to specifically introduce The implementation of whole positioning is achieved by the following way and is described：

S1, positioning region is divided into many sub-regions, multiple elements of a fix points are set in each sub-regions.

Two-dimension square shape grid distribution according to 1m*1m marks off 200 reference points, and adjacent two reference point is in two coordinates Distance on direction of principal axis is 1m.It is a two-dimensional coordinate system with the region, origin is set on the intersection point of region last cell.

Positioning region is divided into 50 positioning subregions by the mode according to 2m*2m, and adjacent two subregion is in two coordinates Distance on direction of principal axis is 2m.It is every sub-regions addition label information 1,2,3..., 50.

S2, terminal gather each coordinate points RSSI finger print informations and coordinate information, by wireless network transmissions to server, Build fingerprint database.

RSSI fingerprints and coordinate information are gathered using Android device successively in 150 reference points, each reference point 10 finger print informations of collection, average.

It is JSON network packets by each reference point collection Information encapsulation, service is sent to by wireless network mode Device, is added in Mysql databases by server.

Server is based on decision tree principle and trains decision tree, determines optimum decision tree depth and leafy node number.This reality It is 6 and 29 to train optimum decision tree depth and leafy node number according to fingerprint database in example.

Above-mentioned steps S1 and S2 is completed in off-line phase, and following steps are completed for on-line stage.

S3, terminal device gather the RSSI fingerprints of point to be determined, and the fingerprint is input into decision tree, and internal junction is carried out successively Point attribution rule judges, until entering leafy node.Point to be determined judges to have navigated to region 18 according to decision tree in this example.

S4, KNN algorithms are used to being matched with classification residing for target, calculate exact position.Take all fingerprints in region 18 As fingerprint to be measured, target sample r={ r₁,...r_N, residing category dataset each sample is designated as { (r_k1,..., r_ki,...,r_km, target sample is calculated with each sample cosine similarity of data set with equation below：

Ascending order is arranged, and filters out preceding K reference point.K values 6 in this example.Drawn using the method based on weighting to be measured Point position coordinates.6 maximum samples of similarity are chosen, is that each coordinate vector defines weight：

Point target positioning result to be measured is as follows：

Wherein, x_kiRepresent i-th coordinate vector abscissa of kth class sample, y_kiRepresent i-th coordinate of kth class sample Vectorial ordinate.

Finally, by follow-up information transmission, coordinate result is returned into positioning terminal and is shown.

So far whole position fixing process is realized.

In sum, the present embodiment performs the side of flow using WiFi location algorithms in the quick KNN rooms based on decision tree Formula comprehensively describes the process positioned in embodiment.The algorithm compared with the WiFi localization methods based on other algorithms, with Under several advantages：Region recognition rate is up to more than 90%；On the positioning trip time, required matching during due to being accurately positioned Fingerprint quantity has been narrowed down in identified region, so the location efficiency of this method is compared to based on global fingerprint matching algorithm Localization method will height；In positioning precision, compared to the KNN algorithms of comparative maturity, positioning precision is higher, and position error can be with It is maintained at 1 to 2m.

Above-described embodiment is the present invention preferably implementation method, but embodiments of the present invention are not by above-described embodiment Limitation, it is other it is any without departing from Spirit Essence of the invention and the change, modification, replacement made under principle, combine, simplification, Equivalent substitute mode is should be, is included within protection scope of the present invention.

Claims (9)

1. WiFi localization methods in a kind of quick KNN rooms based on decision tree, it is characterised in that methods described includes following step Suddenly：

Positioning region is divided into many sub-regions, multiple elements of a fix points are set in each sub-regions；

Terminal gathers each coordinate points RSSI finger print informations and coordinate information, and by wireless network transmissions to server, structure refers to Line database Ψ；

Server is differentiated by integrated decision Tree algorithms to area classification residing for target；

Use KNN algorithms to being matched with classification residing for target, calculate exact position.

2. WiFi localization methods in a kind of quick KNN rooms based on decision tree according to claim 1, described will positioning Region division is many sub-regions, and setting multiple elements of a fix points in each sub-regions specifically includes：

Positioning region is carried out according to dividing mode at equal intervals divide many sub-regions, be that each sub-regions set classification mark Sign；

Arbitrary placement's multiple elements of a fix point, records each point coordinates information in each sub-regions.

3. WiFi localization methods in a kind of quick KNN rooms based on decision tree according to claim 1, described terminal is adopted Collect each coordinate points RSSI finger print informations and coordinate information, wireless network transmissions are passed through after JSON is encapsulated as network packet To server.

4. WiFi localization methods in a kind of quick KNN rooms based on decision tree according to claim 1, it is characterised in that Described server carries out differentiation and specifically includes by integrated decision Tree algorithms to area classification residing for target：

To fingerprint database Ψ and label information, decision tree, the multiple leafy nodes of generation are generated using decision tree training philosophy；

Input target sample enters decision tree root node, carries out rule match with inner branch successively, until target sample enters Leafy node；

Leafy node classification is determined that target sample affiliated area classification is leafy node by inside comprising the most classification of sample number Classification.

5. WiFi localization methods in a kind of quick KNN rooms based on decision tree according to claim 1, it is characterised in that Described use KNN algorithms are matched to classification residing for target, are calculated exact position and are specifically included：

Every cosine similarity of vector in the vectorial fingerprint bases corresponding with residing classification of RSSI of tested point is calculated, by carrying out ascending order Arrangement, takes preceding K reference point and constitutes neighbours' sample set, and the corresponding two-dimensional coordinate of neighbours' sample set constitutes neighbours' sample coordinate collection；

Using the cosine similarity of neighbours' sample set as weight, using the method based on weighting draw tested point position coordinates (x, y)。

6. WiFi localization methods in a kind of quick KNN rooms based on decision tree according to claim 1, it is characterised in that Described fingerprint database Ψ is expressed as：

$\mathrm{\Ψ}=\left[\begin{array}{c}RSS{I}_{1,1},...,RSS{I}_{1,n},...,RSS{I}_{1,N}\\ {\mathrm{RSSI}}_{m,1},...,{\mathrm{RSSI}}_{m,n},...,{\mathrm{RSSI}}_{m,N}\\ {\mathrm{RSSI}}_{M,1},...,{\mathrm{RSSI}}_{M,n},...,{\mathrm{RSSI}}_{M,N}\end{array}\right]$

Wherein RSSI_m,n(m=1,2...M, n=1,2 ... N) represents that the RSSI that m-th reference point receives n-th AP is average Value, each row vector of fingerprint database Ψ represents that a reference point receives the RSSI of N number of AP.

7. WiFi localization methods in a kind of quick KNN rooms based on decision tree according to claim 4, it is characterised in that Described decision tree training philosophy is specifically included：

Regard RSSI vectors as a categorical attribute per one-dimensional component, therefore property set is expressed as：

R (D)={ R₁,...,R_i,...,R_N}

Wherein, R_iRSSI vector i-th dimension components are represented, for RSSI i-th dimension attributes R_i, to these values by sorting from small to large, Obtain ascending sequence { R_i1,...,R_ij,...R_in, set [R_ij,R_ij+1) intermediate pointIt is interval division point, for category Property R_iConstruction candidate divides point set：

$T_R=\left\{\frac{R_{ij}+R_{ij+1}}{2}\right|1\≤i\≤n-1\};$

Structure attribute optimum division point decision rule, i.e. attribute R_iOptimum division point should meet：

$Gain(D,R_i)=\max Ent\left(D\right)-\underset{\mathrm{\λ}\∈\{-,+\}}{\Σ}\frac{D_t^{\mathrm{\λ}}}{D}Ent\left(D_t^{\mathrm{\λ}}\right);$

According to above-mentioned decision rule, optimal dividing point corresponding informance gain is attribute information gain in itself, in construction decision-making During tree, current node attribute should meet：

R=arg max G (D, R_i)

From root node, optimal dividing attribute and optimal dividing point are selected according to above-mentioned rule, by sample set according to division points Two points are carried out for two subsets, is then further divided in the two subsets, until all leafy nodes all include phase Generic sample, completes decision tree and builds.

8. WiFi localization methods in a kind of quick KNN rooms based on decision tree according to claim 5, it is characterised in that The cosine similarity of every vector is specific as follows in the vectorial fingerprint bases corresponding with residing classification of RSSI of described calculating tested point：

Target sample r={ r₁,...r_N, residing category dataset each sample is designated as { (r_k1,...,r_ki,...,r_km, mesh Standard specimen sheet and data set each sample cosine similarity is defined as：

${\mathrm{Sim}}_i=\frac{r_1\×r_{k1}+...+r_N\×r_{kN}}{\sqrt{r_1\×r_1+...+r_N\×r_N}+\sqrt{r_{k1}\×r_{k1}+...+r_{kN}\×r_{kN}}}.$

9. WiFi localization methods in a kind of quick KNN rooms based on decision tree according to claim 5, it is characterised in that The method that described use is based on weighting show that tested point position coordinates (x, y) is as follows：

K maximum sample of similarity is chosen, is that each coordinate vector defines weight：

$w_{ki}=\frac{{\mathrm{Sim}}_i}{\underset{i=1}{\overset{K}{\Σ}}{\mathrm{Sim}}_i},$

Point target positioning result to be measured is as follows：

$\left\{\begin{array}{c}x=\underset{i=1}{\overset{K}{\Σ}}{w}_{ki}{x}_{ki}\\ y=\underset{i=1}{\overset{K}{\Σ}}{w}_{ki}{y}_{ki}\end{array}\right.,$

Wherein, x_kiRepresent i-th coordinate vector abscissa of kth class sample, y_kiRepresent i-th coordinate vector of kth class sample Ordinate.